Classification of Multi-Parametric Body MRI Series Using Deep Learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Multi-parametric magnetic resonance imaging (mpMRI) exams involve diverse series types and protocols.
• Inaccurate DICOM headers in mpMRI data hinder efficient radiologist review.
• Standardizing series identification is crucial for streamlined diagnostic workflows.

Purpose of the Study:

• To develop and evaluate a deep learning model for classifying 8 distinct body mpMRI series types.
• To enhance the efficiency of radiological interpretation by automating series identification.
• To compare the performance of different deep learning architectures for this classification task.

Main Methods:

• Training and comparing ResNet, EfficientNet, and DenseNet classifiers on multi-institutional mpMRI data.
• Evaluating the best-performing model's accuracy with varying training data quantities.
• Assessing model generalization on out-of-distribution and multi-scanner datasets.

Main Results:

• The DenseNet-121 model achieved the highest F1-score (0.966) and accuracy (0.972).
• Accuracy exceeded 0.95 with over 729 training studies, demonstrating scalability.
• The model maintained high accuracy on external datasets (DLDS: 0.872, CPTAC-UCEC: 0.810).

Conclusions:

• The DenseNet-121 deep learning model effectively classifies 8 body mpMRI series types.
• The model demonstrates robust performance across internal and external datasets.
• Automated series classification using deep learning offers a reliable solution for improving radiologist workflow.